Trend Explorer 4 Sb

Trend Explorer


Companies regularly need to create trends for marketed products. There are a number of different ways to trend a product.

The problem is how to know which is the best approach. A poor trending approach will produce a poor trend or a trend that fluctuates from month to month. Either outcome will draw scrutiny from senior management. In addition, this whole process may be cumbersome and time-consuming.

Forecasters would also like to better understand the trending methods and options available. This usually involves experimentation with many different approaches and algorithm parameters.


Companies use several methods to select a trending approach:

Previously used techniques. The forecaster selects data sources, preprocessing, and trending methods that worked well in other situations. The assumption is that previous methods are good enough for the current situation.

Fit to historical data. Sometimes a trending method, for example linear regression or exponential smoothing, is selected because the method fits the historical data well, using a measure such as R-squared or mean absolute percentage error.

Academic theory. A trending method may be selected because it is widely accepted in the statistical forecasting community or has performed well in academic evaluations of forecasting methods.


The goal of a trending method is to predict the eventual sales of a product. Using a goodness-of-fit measure such as R-squared only shows how well the trending method can fit the historical data. It is easy to construct a complicated model that will fit the historical data exactly, yet be completely worthless for trending. Likewise, relying on trending theory or prior experience may result in a better method being overlooked for each particular situation.

Many forecasters use third-party software to calculate product trends. This requires experience with another software program and is not automated, so the process can be tedious.


The Trend Explorer is a tool intended for trend analysis and comparison and uses the Objective Insights Prediction Engine for all trend calculations. Trend Explorer is built in Microsoft Excel, so it is easy to copy data and results to and from other Excel files.

Trend Explorer has two primary functions:

#1: The trend analysis component allows you to produce trends in fully automatic or manual modes. For example, you can select any method and specify custom settings for the method. You can specify the number of data points used for a moving average or regression. Exponential smoothing and Box-Jenkins can be automatically optimized for best results or individual parameters can be changed to see the effect on the trend. The tool provides a number of in-sample trend statistics.

#2: Trend Explorer also offers trend comparison; the results of all 27 trending methods are automatically tested against a series of held-out samples from the end of the provided data. Calculated trends are compared against rolling windows over the length of the held-out sample in order to minimize the effect where different segments of data may favor one method over another. Both the number and length of the rolling windows are selectable. Results of the trend comparison are ranked against several different out-of-sample trend statistics, such as mean absolute percentage error, mean squared error, and Akaike’s Information Criterion (AIC), among others.


Trend Explorer Plus

Objective Insights offers a variant of the Trend Explorer, Trend Explorer Plus, which includes features from the Short-Term Forecaster (STF). Trend Explorer Plus adds the ability to produce a complete product forecast (vs. the Trend Explorer’s focus on trend analysis). Trend Explorer Plus easily imports product data from a central location, runs and tests demand trends, and then shows the results of the best trend and selected comparator trends in terms of units or prescriptions and gross and net revenues. Events such as price increases, gross to net trade discount changes, and inventory changes are incorporated. Both weekly and monthly data versions are available.


Slides 2

For more information, please download the Trend Explorer Overview (pdf).


 Required inputs

Historical monthly, daily, annual, or quarterly sales data (for example, prescriptions or units)

Optional Inputs

Preprocessing quantities such as number of shipping or billing days per month



• Projected sales (for example, prescriptions or units)
• Error ratings (confidence limits) for the trending methods tested
• Mean absolute error (MAE)
• Mean squared error (MSE)
• Mean absolute percentage error (MAPE)
• Symmetric mean absolute percentage error (sMAPE)
• Mean absolute scaled error (MASE)
• R squared/adjusted R squared
• Thiel’s U statistic
• Akaike's information criterion (AIC)
• Schwarz’s Baysian information criterion (BIC)
• Variance of historical data
• Seasonality coefficients for seasonal trending methods